Unverified50% confidenceOpinionExact time
The author himself acknowledges that the model still needs more refinement, and that both the 25,000-sample data scale and the unofficial RLCD method still have room for validation and iteration.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/18/2026
First Seen
Valid until: 12/17/2026
Sources
Related Entities
Related Claims
Unverified2022年Hoffman等人的Chinchilla论文重新校准了算力最优的参数-数据比例，指出此前许多大模型训练不充分66% similarVerifiedIn 2022, Hoffmann et al.'s Chinchilla paper redid experiments on four hundred models and concluded that, under the premise of compute-optimality, each parameter should correspond to approximately 20 tokens.64% similarUnverified25%的风险降幅数字仍需在方法论严格的复制研究中验证，应被视为值得进一步验证的信号而非最终定论62% similarUnverifiedOn a moderately complex prompt (generating a website), the model consumed approximately 20,000 tokens in the thinking process alone.61% similarUnverified该测试样本规模相对有限（5个模型、13个代码库），结论有待更大规模验证61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931576API
curl https://kongchang.com/api/v1/knowledge/claims/931576MCP
get_claim(id=931576)